← Writing
perspion post-mortem

Six users. Zero dollars. I'm shutting down Perspion.

Four months building a product six people used — one of them me. The four things that killed Perspion, and the rules the studio now runs on.

I spent four months building a product that six people used.

One of them was me.

Today I’m shutting it down. This post-mortem is the most useful thing Perspion ever produced.

If you’ve ever shipped a side project, you know the movie.

You pick a platform you genuinely love — for me, YouTube. You spot a real-sounding problem: creators drowning in comments, no idea what their audience actually thinks. You build the thing nights and weekends: TypeScript, React, Supabase, deployed on Vercel. AI analyzes the comments, finds the patterns, hands the creator insights.

It demos beautifully. Your wife thinks it’s clever. You ship it, post about it twice, and wait for the internet to notice.

The internet does not notice.

Here are Perspion’s real numbers, March to July 2026:

  • 6 signups total. Including me. So: five strangers.
  • 4 actually ran an analysis. 2 created an account and never came back.
  • 0 paid. Not one.
  • Last activity: four users in April, one in March, one in July. The July one was me, checking if it still worked.

Six users is not a distribution problem you iterate on. It’s a verdict. But the autopsy found four specific causes of death, and they’re all reusable:

1. I paid for my free users to reject me.

Every free analysis burned AI tokens — my tokens. Three free runs per month meant every curious visitor cost me money and owed me nothing. A free tier should cost you approximately zero per user; mine had a negative unit economy before the pricing page.

2. I was my database’s babysitter.

Supabase’s free tier pauses your database after a week of inactivity. A product with six users is, by definition, inactive. So I had a recurring chore: log in and poke my own infrastructure so it wouldn’t fall asleep. When keeping the product technically alive requires manual labor, the product is telling you something.

Vercel’s Hobby plan prohibits commercial use. If someone had actually paid me, my hosting would have violated the terms of service on day one. I had built a business on infrastructure that contractually forbade business. Nobody checks this until it matters — check it before it matters.

4. I built for a platform I love, not a pain someone would pay to kill.

I love YouTube. That’s why I built this — and that’s the trap. “I love this platform” is not “creators will pay monthly to analyze comments.” I never validated pull; I shipped a solution and went looking for its problem. And then I barely marketed it, because deep down I was waiting for the product to sell itself. Products don’t.

The meta-lesson underneath all four: every one of these was knowable before I wrote a line of code. Unit economics, infra terms, time-to-value, distribution plan — none of this requires building. It requires honesty at the whiteboard.

Conclusion

So Perspion is dead. The database can finally sleep.

But the studio it leaves behind now runs on rules written in its blood:

  • Zero marginal cost per free user. Not because subsidy is always wrong, but because a solo studio with no funding cannot buy customers it has no evidence it will keep. If a free visitor costs me money before anyone has ever paid, the design is wrong.
  • Wow in under 10 seconds — no product that requires weeks of habit before it gets good.
  • Every output is a shareable artifact. Distribution designed in, not bolted on.
  • Kill criteria written before launch. Numbers decide, not attachment. Perspion lived four months too long because no number was ever allowed to judge it.

Two products built under those rules are already live or on the calendar: a daily game about what things used to cost, and something for NFL fans that turns your suffering into a score — launching kickoff week.

Six users taught me more than a thousand would have. Watch what I do with the tuition.